Extensive soil surface observations and measurements were conducted in Normandy on loess soils prone to surface crusting in order to understand the spatial distribution and temporal variability of soil surface characteristics involved in erosion processes. A database composed of more than 5000 soil surface observations at the field or plot scale was analysed. We classified these observations according to combinations of three dominant factors: crusting stage, roughness and crop cover. Frequency analysis showed that three combinations of these factors accounted for more than 50% of the total observations. Eight other combinations accounted each for more than 2% of the observations. These dominant and secondary combinations are considered as “reference surface conditions”. Their potential range of the infiltration capacity and sediment production were estimated from field measurements under natural and simulated rainfall. For the loess soils in our study, mean field infiltration capacity ranged between 2 mm/h and 50 mm/h while mean sediment concentration ranged between 0.5 g/l and 30 g/l for the different combinations of soil surface and rainfall characteristics. This work, based on a local case study, allowed the development of a new approach for erosion prediction modelling (STREAM) which could be transferable to other areas where Hortonian processes are dominant.